v2.14.0

Hey everyone, the wait is over! Stability Matrix v2.14.0 has officially landed! ✨

After a packed development cycle (thanks for your patience!), we're thrilled to roll out a massive update with new features, integrations, performance boosts, and important fixes based on your feedback.

🪄 Introducing: Prompt Amplifier

Need inspiration or want to add more depth to your prompts? Click the magic wand 🪄 in the Inference prompt editor to use the Prompt Amplifier! Our AI assistant, powered by our experimental Spark model, will then generate creative expansions of your prompt.

⚡ Accelerated Model Discovery (Beta for Insider+)

Tired of waiting for search results? We hear you! Accelerated Model Discovery utilizes our optimized server connections to dramatically speed up browsing online repositories like Civitai. This optional connected enhancement was made possible by community support – local features will always remain free and open-source.

model_discovery.mp4

✨ More v2.14.0 Highlights: ✨

• New Packages: Welcome Stable Diffusion WebUI AMDGPU Forge and Forge - Classic! Plus, easily install Triton & SageAttention for ComfyUI.

• Enhanced Inference: Added HiDream and Wan 2.1 T2V/T2I video projects support, base model filtering for Extra Networks, Rescale CFG, Align Your Steps scheduler, wildcard support ({a|b|c}), Plasma Noise, and a handy width/height swap button.

• 

• Smarter Prompt Editing: AST parsing improves prompt understanding, plus Ctrl+Up/Down hotkeys to easily adjust token weights!

• Expanded Model Repos: Added an OpenModelDB tab and Wan 2.1 files to the HuggingFace browser.

• Compatibility & Bug Fixes: Upgraded HIP SDK, fixed issues with RTX 5000, macOS file handling & display quirks, package updates, model metadata, image caching, and much more!

A massive thank you to our incredible Patreon Supporters! Your support fuels the continued development and maintenance of Stability Matrix. We couldn't do this without you!

Full Changelog

Added

New Packages & Integrations

• Added new package - Stable Diffusion WebUI AMDGPU Forge
• Added new package - Stable Diffusion WebUI Forge - Classic
• Added new Package Command (in the 3-dots menu) for installing Triton & SageAttention in ComfyUI

Inference Features

• Added Prompt Amplifier to Inference - click the magic wand 🪄 in the prompt editor to expand and enrich your ideas. Tailor the vibe with the ‘Feel’ selector and watch as your generations come to life with extra detail, coherence, and flair!
• Added support for HiDream in Inference - see ComfyUI Examples for more details
• Added toggle for filtering Inference Extra Networks by base model
• Added Rescale CFG addon to Inference
• Added Swap Dimensions button between the width/height input in Inference
• Added Ctrl+Tab/Ctrl+Shift+Tab shortcuts for navigating between Inference tabs
• Added Align Your Steps scheduler to Inference
• Added wildcards to Inference prompts, e.g. {blue|green|red} will randomly select one of the colors
• Added Wan 2.1 Text to Video and Text to Image project types for Inference
• Added new autocomplete tag source to Inference - Danbooru/e621 merged tags
• Added Abstract Syntax Tree (AST) parsing for Inference prompts. This provides a more robust internal understanding of prompt structure, paving the way for future enhancements.
• Added hotkey (Ctrl+Up/Ctrl+Down) in Inference prompt editors to adjust the weight emphasis of the token under the caret or the currently selected text.
  • This automatically wraps the token/selection in parentheses () if it's not already weighted.
  • It modifies existing weights within parentheses or adds weights if none exist (e.g. (word:1.1)).
  • Handles selection spanning multiple tokens intelligently.
• Added Plasma Noise addon to Inference for text to image workflows

Model Management & Discovery

• Added Accelerated Model Discovery (Beta) (⚡ icon in Civitai Browser) for Insider+ supporters. Utilizes an optimized connection for dramatically faster, more responsive browsing of online model repositories.
• Added OpenModelDB tab to the Model Browser
• Added Wan 2.1 files to the HuggingFace model browser

User Interface & Experience (UI/UX)

• Added Undo/Redo commands to text editor context menus

Internal / Developer Changes

• (Internal) Introduced unified strategy pattern (IConfigSharingStrategy) to for handling different config file formats (JSON, YAML, FDS).
  • Added support for configuring nested paths in JSON and YAML files (e.g. paths.models.vae) via dot-notation in SharedFolderLayoutRule.ConfigDocumentPaths.
  • Packages can now use the SharedFolderLayout property to define a ConfigFileType and ConfigSharingOptions (like RootKey), without needing to implement custom configuration logic.

Changed

Inference Features

• Improved the quality of Inference inpainting by upgrading the workflow behind the scenes. The workflow remains the same for you — just better results!
• FaceDetailers in Inference will now inherit the primary sampler/scheduler/etc. by default. You can still manually set these by enabling the options via the ⚙️ button on the FaceDetailer card
• Slightly rearranged the FaceDetailer card layout due to the above change
• Inference "Extra Networks" selector now filters extra networks based on the selected base model
• Merged Inference GGUF workflows into the UNet model loader option (no longer need to choose GGUF separately)

Model Management & Discovery

• Changed the names of some of the shared model folders to better reflect their contents
• Improved Checkpoint Manager memory usage (thanks to @firegeek for the profiling assistance!)
• Performance optimizations for Checkpoint Manager (progress indicators now fully uses Compiled Bindings)

Package Management & Compatibility

• Upgraded HIP SDK installs to 6.2.4 for ComfyUI-Zluda and AMDGPU-Forge
• Updated install for kohya_ss to support RTX 5000-series GPUs

User Interface & Experience (UI/UX)

• Improved window state handling
• Updated some date strings to take into account the user's locale

Localization

• Updated Japanese, Brazilian Portuguese, Chinese, and Russian translations

Internal / Developer Changes

• (Internal) Upgraded FluentAvalonia to 2.3.0
• (Internal) Refactored configuration-based shared folder logic: Centralized handling into SharedFoldersConfigHelper and format-specific strategies, removing custom file I/O logic from individual package classes for improved consistency and maintainability.
  • Migrated packages ComfyUI (incl. Zluda), VladAutomatic (SD.Next), Sdfx, and StableSwarm to use the unified system for configuration and symlink based sharing.

Fixed

Installation, Compatibility & Core Functionality

• Fixed RTX 5000-series GPU detection in certain cases
• Fixed Package Updates and Change Version not using stored PyTorch index and instead using the default recommended index.
• Fixed ComfyUI-Zluda not being recognized as an option for Inference or SwarmUI (for real this time)
• Fixed errors from invalid pip specifiers in requirements files

User Interface & Experience (UI/UX)

• Fixed Image Viewer animation loader keeping file handles open, which resolves 2 different issues (OSes are fun):
  • (macOS) Fixed FileNotFoundException crash when navigating to Inference tab after deleting a Webp animation file previously opened in the Image Viewer Dialog.
  • (Windows) Fixed Webp animation files unable to be deleted without closing the app first.
• Fixed Image Viewer FileNotFoundException on fetching image size, if navigating to a deleted image file.
• (macOS) Fixed Webp / Gif animations RGB colors flipped.
• Fixed window disappearing on macOS when the saved window size is very small
• Fixed large white boxes appearing when tooltips are visible on macOS/Linux
• Fixed package images sometimes showing as blank due to concurrent image caching. Requests to same image resources are now de-duplicated
• Reduced memory usage from ShowDisabledTooltipExtension

Inference & Workflows

• Fixed some cases of missing custom nodes in SwarmUI
• Fixed Inference ControlNet Preprocessors using incorrect resolution and increased maximum of smallest dimension to 16384
• Fixed Inference Extra Networks card not allowing for more than one model at a time

Model Management & Discovery

• Fixed missing base model options in the Metadata Editor
• Fixed some crashes when using Accelerated Model Discovery

Supporters

Visionaries

Our heartfelt gratitude goes out to our amazing Visionary-tier Patrons: Waterclouds, Corey T, bluepopsicle, Bob S, Akiro_Senkai, and Ibixat! Your exceptional support is fundamental to the ongoing development and success of Stability Matrix. We are immensely grateful for your partnership and belief in the project! 🙏

Pioneers

We also want to give a huge thank you to our dedicated Pioneer-tier Patrons: tankfox, Mr. Unknown, Szir777, Tigon, NowFallenAngel, Al Gorithm, and welcome to our newest Pioneer, Noah M.! Your consistent support and enthusiasm keep the momentum going. Thank you all for being such an important part of our community! ✨